If you find the content useful and wish to support our platform’s development, you can contribute any amount toward our production costs. Scan the UPI QR code for payment within India. Or use the Ko-fi link to process a secure payment via PayPal.
Machine Learning
26 Terms
A–Z Linked
Machine Learning
Terms related to how machines learn patterns from data and make predictions or decisions.
Select a term below or scroll through the full category.
- Accuracy
- Bias
- Clustering
- Cross-Entropy Loss
- Entropy
- Feature
- Feature Engineering
- Generalisation
- Hyperparameter
- Inductive Bias
- Information Theory
- Knowledge Distillation
- Learning Rate
- Loss Landscape
- Machine Learning
- Model Evaluation
- Optimisation
- Overfitting
- Regression
- Reinforcement Learning
- Supervised Learning
- Test Data
- Training Data
- Underfitting
- Unsupervised Learning
- Validation Data
Accuracy
📘 Definition
Accuracy measures the proportion of correct predictions made by a model.
🪄 Simple Explanation
Accuracy tells how often a model is correct.
🔑 Why It Matters
It is one of several important evaluation metrics.
💡 Example
A classifier correctly predicts 95 out of 100 cases.
See Also
- Model Evaluation
- Precision
- Recall
Bias
📘 Definition
Bias is a systematic error that causes a machine learning model to make consistently inaccurate or unfair predictions because of assumptions in the model or the data.
🪄 Simple Explanation
Bias means a model tends to make certain mistakes repeatedly.
🔑 Why It Matters
Recognising bias helps build fairer and more reliable AI systems.
💡 Example
A hiring model trained on unbalanced historical data may unfairly favour one group.
See Also
Clustering
📘 Definition
Clustering is an unsupervised learning technique that groups similar data points without predefined labels.
🪄 Simple Explanation
Clustering discovers natural groups in data.
🔑 Why It Matters
It helps reveal hidden patterns and customer segments.
💡 Example
A retailer groups customers by purchasing behaviour.
See Also
- Unsupervised Learning
- Dataset
- Pattern
Cross-Entropy Loss
📘 Definition
Cross-entropy loss is a loss function that measures how closely a model's predicted probability distribution matches the correct outcome.
🪄 Simple Explanation
Cross-entropy loss measures how far a model's predicted probabilities are from the correct answer.
🔑 Why It Matters
It is widely used to train classification and language models because it provides a numerical signal for improving probability predictions.
💡 Example
If the correct class is 'cat' but a classifier gives 'cat' only a very small probability, the cross-entropy loss will be high.
See Also
- Loss Function
- Probability
- Optimisation
Continue Learning
Entropy
📘 Definition
Entropy is a measure of uncertainty in a probability distribution. A distribution with several similarly likely outcomes has higher entropy than one dominated by a single highly probable outcome.
🪄 Simple Explanation
Entropy measures how uncertain a set of possible outcomes is.
🔑 Why It Matters
Entropy is foundational to information theory and appears in important machine learning objectives used by classification systems and language models.
💡 Example
If a classifier assigns almost equal probabilities to several classes, its prediction has relatively high entropy.
See Also
- Probability
- Loss Function
- Prediction
Continue Learning
Feature
📘 Definition
A feature is an individual measurable characteristic or variable used by a machine learning model to make predictions.
🪄 Simple Explanation
Features are pieces of information the model learns from.
🔑 Why It Matters
Choosing good features often improves model performance.
💡 Example
House size and location are features in a house-price prediction model.
See Also
Feature Engineering
📘 Definition
Feature engineering is the process of selecting, creating, or transforming variables so that a machine learning model can learn more effectively.
🪄 Simple Explanation
It prepares data in a way that helps AI learn better.
🔑 Why It Matters
Well-designed features often improve accuracy more than changing algorithms.
💡 Example
Combining day and month into a seasonal feature may improve sales prediction.
Generalisation
📘 Definition
Generalisation is a model's ability to perform well on new, unseen data rather than only on the data used for training.
🪄 Simple Explanation
A good model should work well on new examples.
🔑 Why It Matters
Generalisation is one of the main goals of machine learning.
💡 Example
A handwriting model correctly recognises letters written by people it has never seen before.
See Also
Hyperparameter
📘 Definition
A hyperparameter is a setting chosen before training begins that influences how a machine learning model learns from data.
🪄 Simple Explanation
Hyperparameters control how a model learns.
🔑 Why It Matters
Choosing suitable hyperparameters can significantly improve model performance.
💡 Example
Changing the learning rate before training is adjusting a hyperparameter.
See Also
Inductive Bias
📘 Definition
Inductive bias refers to the assumptions built into a machine learning algorithm or model architecture that influence the patterns it is more likely to learn from data.
🪄 Simple Explanation
Inductive bias is the set of built-in assumptions that guides what a model learns easily.
🔑 Why It Matters
Different inductive biases make particular architectures better suited to particular problems and help explain why model design affects learning efficiency and generalisation.
💡 Example
A convolutional neural network contains assumptions that make it particularly effective at recognising local patterns in images.
See Also
- Model
- Generalisation
- Neural Network
Continue Learning
Information Theory
📘 Definition
Information theory is a mathematical framework for measuring information, uncertainty, and the efficient representation and communication of data.
🪄 Simple Explanation
Information theory provides mathematical ways to measure how much information or uncertainty is present in data.
🔑 Why It Matters
Its concepts help explain how models learn from data, measure prediction uncertainty, compress information, and compare predicted probabilities with correct outcomes.
💡 Example
A language model reduces uncertainty about the next token by assigning different probabilities to possible continuations.
See Also
- Probability
- Loss Function
- Language Model
Continue Learning
Knowledge Distillation
📘 Definition
Knowledge distillation is a technique in which a smaller model, called the student, is trained to reproduce useful behaviour learned by a larger or more capable teacher model.
🪄 Simple Explanation
Knowledge distillation teaches a smaller model to imitate a larger model.
🔑 Why It Matters
Distillation can preserve much of a powerful model's capability while reducing memory use, computing requirements, latency, and deployment cost.
💡 Example
A large language model may act as a teacher while a smaller model learns from its outputs so the resulting system can run more efficiently.
Learning Rate
📘 Definition
The learning rate determines how much a model changes its internal parameters after each learning step during training.
🪄 Simple Explanation
The learning rate controls how quickly a model learns.
🔑 Why It Matters
A suitable learning rate improves both speed and accuracy.
💡 Example
A very high learning rate may prevent a model from reaching the best solution.
See Also
Loss Landscape
📘 Definition
A loss landscape is a conceptual map showing how a model's loss changes as its parameters change during training.
🪄 Simple Explanation
The loss landscape shows possible model settings and how much error each one produces.
🔑 Why It Matters
Training involves navigating this landscape towards parameter values that reduce loss, making the concept useful for understanding optimisation and generalisation.
💡 Example
Gradient descent moves a model through the loss landscape towards settings that reduce prediction error.
See Also
- Loss Function
- Gradient Descent
- Optimisation
Continue Learning
Machine Learning
📘 Definition
Machine learning is a branch of artificial intelligence in which computer systems learn patterns from data instead of being programmed with every rule manually. These systems use examples, features, and feedback to make predictions, classify information, detect patterns, or improve performance over time.
🪄 Simple Explanation
Machine learning allows computers to learn from examples and use that learning to make decisions or predictions.
🔑 Why It Matters
Machine learning powers many everyday AI applications, including recommendations, fraud detection, medical prediction, search engines, and language tools. It is one of the most important foundations for understanding modern AI.
💡 Example
An email system that learns to identify spam by studying thousands of previous spam and non-spam messages is using machine learning.
Model Evaluation
📘 Definition
Model evaluation is the process of measuring how well a machine learning model performs using appropriate metrics and datasets.
🪄 Simple Explanation
Evaluation checks whether a model is reliable.
🔑 Why It Matters
Good evaluation ensures AI systems are trustworthy before deployment.
💡 Example
A team compares two models using accuracy and recall before release.
See Also
Optimisation
📘 Definition
Optimisation is the process of adjusting a model so that prediction errors are reduced and overall performance improves.
🪄 Simple Explanation
Optimisation helps AI models become more accurate.
🔑 Why It Matters
Nearly every machine learning model relies on optimisation techniques.
💡 Example
Training repeatedly updates a model until prediction errors become smaller.
See Also
Overfitting
📘 Definition
Overfitting occurs when a model learns the training data too closely, including noise, and performs poorly on new data.
🪄 Simple Explanation
An overfitted model memorises instead of generalising.
🔑 Why It Matters
Avoiding overfitting improves reliability.
💡 Example
A model scores 99% on training data but only 70% on new data.
See Also
Regression
📘 Definition
Regression is a supervised learning technique used to predict continuous numerical values rather than categories.
🪄 Simple Explanation
Regression predicts numbers instead of labels.
🔑 Why It Matters
Many forecasting problems rely on regression.
💡 Example
Predicting tomorrow's temperature is a regression task.
See Also
Reinforcement Learning
📘 Definition
Reinforcement learning is a machine learning approach in which an agent learns by taking actions and receiving rewards or penalties. Over time, the agent learns which actions lead to better outcomes in a particular environment.
🪄 Simple Explanation
Reinforcement learning teaches an AI system through trial and error, using rewards to encourage better behaviour.
🔑 Why It Matters
Reinforcement learning is important for tasks involving sequences of decisions, such as robotics, game playing, autonomous driving, resource allocation, and optimisation problems.
💡 Example
A game-playing AI can learn better strategies by playing many rounds, receiving positive rewards for winning moves and negative feedback for poor choices.
See Also
- Agent
- Machine Learning
- Reward Function
- Optimisation
Supervised Learning
📘 Definition
Supervised learning is a machine learning approach in which a model is trained using labelled examples. Each training example includes both input data and the correct output, allowing the model to learn the relationship between them and apply that learning to new cases.
🪄 Simple Explanation
Supervised learning teaches a model using examples where the correct answer is already known.
🔑 Why It Matters
Many practical AI tasks depend on supervised learning because organisations often want systems that can predict known outcomes, such as risk level, customer category, disease status, or exam performance.
💡 Example
A bank may train a supervised learning model using past loan applications labelled as “repaid” or “defaulted” so the model can estimate risk for new applicants.
Test Data
📘 Definition
Test data is a separate dataset used only after training is complete to measure how well a model performs on unseen data.
🪄 Simple Explanation
Test data measures how well an AI model performs on new information.
🔑 Why It Matters
Reliable evaluation requires data the model has never seen before.
💡 Example
Researchers report model accuracy using a reserved test dataset.
See Also
Training Data
📘 Definition
Training data is the data used to teach a machine learning model. It contains examples from which the model learns patterns, relationships, categories, or behaviours that it can later apply to new information.
🪄 Simple Explanation
Training data is the set of examples an AI model studies while learning how to perform a task.
🔑 Why It Matters
The quality, relevance, and balance of training data strongly influence the performance of an AI system. Poor training data can lead to inaccurate, biased, or unreliable results.
💡 Example
To train an AI system to recognise handwritten digits, developers may provide thousands of labelled images showing different ways people write numbers from 0 to 9.
See Also
Underfitting
📘 Definition
Underfitting occurs when a model is too simple to capture important patterns in the data.
🪄 Simple Explanation
An underfitted model has not learned enough.
🔑 Why It Matters
It leads to poor performance on both training and new data.
💡 Example
Using a very simple model for complex medical diagnosis may underfit.
See Also
Unsupervised Learning
📘 Definition
Unsupervised learning is a machine learning approach in which a model analyses data without being given correct answers or labels. The goal is usually to discover hidden patterns, groups, structures, or relationships within the data.
🪄 Simple Explanation
Unsupervised learning looks for patterns in data when the correct answers have not been provided in advance.
🔑 Why It Matters
Unsupervised learning is useful when organisations have large amounts of data but do not yet know what patterns exist. It supports customer segmentation, anomaly detection, topic discovery, and exploratory analysis.
💡 Example
An online retailer may use unsupervised learning to group customers with similar browsing and buying behaviour, even if those groups were not defined beforehand.
See Also
- Machine Learning
- Clustering
- Dimensionality Reduction
- Anomaly Detection
Validation Data
📘 Definition
Validation data is used during model development to tune parameters and compare alternative models before final testing.
🪄 Simple Explanation
Validation data helps developers improve a model before it is finalised.
🔑 Why It Matters
It reduces overfitting and supports better model selection.
💡 Example
A data science team uses validation data to choose the best-performing fraud detection model.